Basic Processing Methods of 3D Geometrical Signals from Large-Scale Scenes

An engineering approach for reconstructing large scenes from laser range data is presented. 3D structure features are first extracted from the range data and false features are recognized to simplify the dense data-sets and provide stable features for registration. Each two overlapping range images are automatically registered using feature units and minimal spanning tree is employed to realize global registration of multiple range images. Finally multi-resolution meshes of scenes are built and local blending is applied to fill holes. Results for building large-scale 3D models of historic sites and urban structures are presented; showing the robustness of these methods.